講演抄録/キーワード |
講演名 |
2022-12-14 15:55
A study on intra-modal constraint loss toward cross-modal recipe retrieval ○Jiahang Lu(Nagoya Univ.)・Haruya Kyutoku(Aichi Univ. of Technology)・Keisuke Doman(Chukyo Univ.)・Takahiro Komamizu(Nagoya Univ.)・Yasutomo Kawanishi(RIKEN)・Takatsugu Hirayama(Univ. of Human Environments)・Ichiro Ide(Nagoya Univ.) |
抄録 |
(和) |
Recipe retrieval has become a popular task in multimedia research due to the importance of food in normal lives. With the development of neural networks, the limit for this task is no longer in the encoder of image or text but learning a better embedding space across image-text modalities. In this work, we propose the usage of an intra-modal constraint loss function for learning the joint embedding of image and text. In this way, the violation of negative pairs from the same homogeneous modality could be reduced. We investigate the effectiveness of the proposed method through experiments on a cooking recipe dataset Recipe1M. |
(英) |
Recipe retrieval has become a popular task in multimedia research due to the importance of food in normal lives. With the development of neural networks, the limit for this task is no longer in the encoder of image or text but learning a better embedding space across image-text modalities. In this work, we propose the usage of an intra-modal constraint loss function for learning the joint embedding of image and text. In this way, the violation of negative pairs from the same homogeneous modality could be reduced. We investigate the effectiveness of the proposed method through experiments on a cooking recipe dataset Recipe1M. |
キーワード |
(和) |
/ / / / / / / |
(英) |
Recipe retrieval / Multimedia / Intra-modal constraint loss function / / / / / |
文献情報 |
信学技報 |
資料番号 |
|
発行日 |
|
ISSN |
|
PDFダウンロード |
|